On Mon, Aug 08, 2016 at 12:10:00PM +0100, Bin.Cheng wrote:
> On Mon, Aug 8, 2016 at 11:40 AM, James Greenhalgh
> <james.greenha...@arm.com> wrote:
> > On Mon, Aug 01, 2016 at 01:18:53PM +0000, Bin Cheng wrote:
> >> Hi,
> >> This is the 3rd version patch implementing vcond_mask and vec_cmp patterns
> >> on
> >> AArch64. Bootstrap and test along with next patch on AArch64, is it OK?
> >
> > OK, with a couple of comments below, one on an extension and once style nit.
> >
> >> 2016-07-28 Alan Lawrence <alan.lawre...@arm.com>
> >> Renlin Li <renlin...@arm.com>
> >> Bin Cheng <bin.ch...@arm.com>
> >>
> >> * config/aarch64/aarch64-simd.md (vec_cmp<mode><mode>): New pattern.
> >> (vec_cmp<mode><v_cmp_result>): New pattern.
> >> (vec_cmpu<mode><mode>): New pattern.
> >> (vcond_mask_<mode><v_cmp_result>): New pattern.
> >
> >> +(define_expand "vcond_mask_<mode><v_cmp_result>"
> >> + [(match_operand:VALLDI 0 "register_operand")
> >> + (match_operand:VALLDI 1 "nonmemory_operand")
> >> + (match_operand:VALLDI 2 "nonmemory_operand")
> >> + (match_operand:<V_cmp_result> 3 "register_operand")]
> >> + "TARGET_SIMD"
> >> +{
> >> + /* If we have (a = (P) ? -1 : 0);
> >> + Then we can simply move the generated mask (result must be int). */
> >> + if (operands[1] == CONSTM1_RTX (<MODE>mode)
> >> + && operands[2] == CONST0_RTX (<MODE>mode))
> >> + emit_move_insn (operands[0], operands[3]);
> >> + /* Similarly, (a = (P) ? 0 : -1) is just inverting the generated mask.
> >> */
> >> + else if (operands[1] == CONST0_RTX (<MODE>mode)
> >> + && operands[2] == CONSTM1_RTX (<MODE>mode))
> >> + emit_insn (gen_one_cmpl<v_cmp_result>2 (operands[0], operands[3]));
> Hi,
> Thanks for reviewing, here is updated patch.
> >
> > Should we be also be catching these in a generic way before expanding?
> This pattern was copied unchanged from Alan's original patch.
> Yes, standard match-and-simplify pattern may be added to check such
> special simplifications. Not sure what the current behavior is, we'd
> better keep this at expanding too?
Yes, that's what I'd expect.
> Comment issues fixed. Is this version OK?
Yes this is OK.
Thanks,
James
